SDXL version of the Sai Pallavi LoRA
Description
FAQ
Comments (6)
Hi, your work is amazing. Keep up the good work.
What learning rate do you use with constant optimizer.
Your settings in lora says None ...how do you get "None" learning rate.
Thanks
@Kharbanda Thank you! Since i use Adafactor, it adapts the learning rate as it goes along. I set the LR, TE, and ULR at 1 and then watch the loss rate. I usually pick the epoch where the loss rate is between .09 and 0.1. Hope that helps!
@Desi_Cafe Thanks...where to find regulization images you use?
Also when I try to use adafactor as scheduler and optimizer both, it gives me list of index error..any idea why..
Thanks once again
@Kharbanda I don't use regularization images anymore, for celebrity LoRAs, and especially for SDXL based models. It slows down the process and at times introduces artifacts that don't sync up with the celeb's face.
On your error - you're best off looking through open and closed issues on the repo. That's usually where I find the best support.
@Desi_Cafe thanks a lot once again....pls try to make lora for kriti sanon..its a request... thanks
bro, attach image dataset with it